In parallel computers (for example, a supercomputer) that are jointly used by a plurality of users, if job execution instructions from the users are concentrated in a specific time slot, computational resources become insufficient, and thus it becomes difficult to execute jobs. Thus, in related-art techniques, a scheduler is provided in a system that manages a parallel computer, and a job execution schedule is managed by the scheduler in order to avoid a shortage of computational resources during the execution of jobs.
However, a job that is not scheduled for execution in advance, but is to be urgently executed sometimes occurs in a real operation. For example, if a disaster occurs, a user has to urgently perform a job for predicting the impact of the disaster in order to take measures against the disaster.
In the above-described technique, if the scheduler receives a new job to be executed, the scheduler identifies a CPU having a free memory space larger than the amount of memory used by the job from a plurality of CPUs, and causes the identified CPU to execute the job. However, in this method, a free memory space is managed for each CPU, and thus scheduling is sometimes performed such that the computing power of the CPUs is not fully utilized.
As an example of a related-art technical document, Japanese Laid-open Patent Publication No. 7-249010 is known.